DTIC ADA461213: Image-Based Techniques for Digitizing Environments and Artifacts

9 Pages·3.5 MB·English

by  Defense Technical Information Center

Image-Based Techniques for Digitizing Environments and Artifacts
PaulDebevec
UniversityofSouthernCaliforniaInstituteforCreativeTechnologies
GraphicsLaboratory
13274FijiWay,5thfloor,MarinadelRey,California,USA,90292
[email protected]
Abstract artifact’ssurfacegeometryanditsphotometricproperties-
theinformationnecessarytorenderitfromdifferentview-
This paper presents an overview of techniques for gen- ing angles and seen under various forms of illumination.
erating photoreal computer graphics models of real-world Despitemuchinsightfulworkinthisarea,nogeneralfully
places and objects. Our group’s early efforts in model- technique exists that for faithfully digitizing objects hav-
ing scenes involved the development of Fac¸ade, an inter- ing arbitrarily complex geometry and spatially-varying re-
activephotogrammetricmodelingsystemthatusesgeomet- flectanceproperties.
ric primitives to model the scene, and projective texture The computer graphics research my group has under-
mappingtoproducethesceneappearanceproperties. Sub- taken in recent years has involved a variety of object and
sequent work has produced techniques to model the inci- scene modeling methods, ranging from techniques based
dentilluminationwithinscenes,whichwehaveshowntobe principallyonrecoveringgeometrytothosebasedmoresig-
useful for realistically adding computer-generated objects nificantlyonacquiringimages. InthispaperIwilldiscuss
to image-based models. More recently, our work has fo- severaloftheseprojectswithaparticularregardtothena-
cussedonrecoveringlighting-independentmodelsofscenes tureofthemodelsacquired,whythemodelswereappropri-
andobjects,capturinghoweachpointonanobjectreflects atefortheparticularapplicationsbeingexplored,andwhich
light. Our latest work combines three-dimensional range aspects of the artifacts have been left unrecorded. While
scans,digitalphotographs,andincidentilluminationmea- noneoftheseprojectsonitsownpresentsageneralsolution
surementstoproducelighting-independentmodelsofcom- tothesceneandobjectdigitizingproblem,evaluatingtheir
plexobjectsandenvironments. strengthsandweaknessestogethersuggestsnewavenuesfor
combininggeometricandphotometricmeasurementsofob-
jectsfordigitizingamoregeneralsetofcomplexobjects.
1.Introduction
2. Immersion ’94: Modeling an Environment
fromStereo
A basic goal of three-dimensional modeling and imag-
inghasbeentodigitizephotorealisticmodelsofreal-world
places and objects. The potential benefits of these tech- In1994IworkedinagroupledbymediaartistMichael
niques to cultural heritage are significant: objects and Naimarkworkingtoanalyzeasequenceofforward-looking
places can be archived in digital forms less susceptible to stereoimageshehadshotatone-meterintervalsalongsev-
decayandplunder;scholarscantestandcommunicatethe- eralmilesofthetrailsinBanffNationalForest.Ourgoalfor
oriesofreconstructionwithoutdisturbingoriginalartifacts; thesummerwastoturnthesesetsofstereoimagepairsinto
and the world’s artistic and architectural heritage can be aphotorealisticvirtualenvironment.Thetechniqueweused
vividly and widely shared among all those with access to was to determine stereo correspondences, and thus depth,
a computer. It will likely be a long time before an exact between left-right pairs of images, and then to project the
computerrepresentationofanartifact–arecord,onecould corresponded pixels forward into the 3D world. For this
suppose,oftheobject’satomicstructure–willbeabletobe weusedastereoalgorithmdevelopedbyJohnWoodilland
nondestructively recorded, stored, and rendered. Instead, Ramin Zabih [18]. To create virtual renderings, we pro-
most digitizing techniques have focussed on recording an jected a supersampled version of the points onto a virtual
Report Documentation Page Form Approved
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.
1. REPORT DATE 3. DATES COVERED
2003 2. REPORT TYPE 00-00-2003 to 00-00-2003
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Image-Based Techniques for Digitizing Environments and Artifacts
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
University of California,Institute for Creative Technologies,13274 Fiji REPORT NUMBER
Way,Marina del Rey,CA,90292
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
The original document contains color images.
14. ABSTRACT
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON

a. REPORT b. ABSTRACT c. THIS PAGE 8
unclassified unclassified unclassified
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
image plane displaced from the original point of view, us-
ingaZ-buffertoresolvetheocclusions.Withjustonestereo
pair, we could realistically re-render the scene from any-
where up to a meter away from the original camera po-
sitions, except for artifacts resulting from areas that were
unseen in the original images, such as the areas originally
hiddenbehindtreetrunks.Tofillintheseoccludedareasfor
novel views, our system would pick the two closest stereo
pairs to the desired virtual point of view, and render both
tothedesirednovelpointofview. Theseimageswerethen
opticallycompositedsothatwhereveronelackedinforma-
tion, the other would fill it in. In areas where both images
hadinformation,thedatawaslinearlyblendedaccordingto
which original view the novel view was closer to. The re-
sultwastheabilitytorealisticallymovethroughtheforest,
aslongasonekeptthevirtualviewpointwithinaboutame-
teroftheoriginalpath. Naimarkpresentedthisworkatthe
SIGGRAPH95panel“MuseumswithoutWalls: NewMe-
diaforNewMuseums”[1],andtheanimationsmaybeseen
attheImmersionprojectwebsite[15].
Theresultsthatwereachievedwereexcitingatthetime;
a scene that at the beginning seemed impossible to model
hadbeen”digitized”inthesensethatnearlyphotorealren-
dered images could be created of flying down the forest
trail. Nonetheless, the model of the scene obtained by the Figure1.TheImmersion’94image-basedmodel-
technique did not allow the virtual camera to move more ing and rendering project. The top images are
than a meter or so from the original path, and the scene a stereo pair (reversed for cross-eyed stereo
could only be seen in the original lighting in which it was viewing) taken in Banff National Forest. The
photographed. ThenextprojectIwilldiscussaimedtofree middle left photo is a stereo disparity map
thecameratoflyaroundthescene. produced by John Woodfill’s parallel imple-
mentation of the Zabih-Woodfill stereo algo-
3 Photogrammetric Modeling and Render- rithm[18]. Toitsrightthemaphasbeenpro-
ingwithFac¸ade cessed using a left-right consistency check
to invalidate regions where running stereo
based on the left image and stereo based
My thesis work [6] at Berkeley done in collaboration
on the right image did not produce consis-
with C.J. Taylor presented a system for modeling and ren-
tentresults. Belowaretwovirtualviewsgen-
deringarchitecturalscenesfromphotographs.Architectural
erated by casting each pixel out into space
scenesareaninterestingcaseofthegeneralmodelingprob-
based on its computed depth estimate, and
lem since their geometry is typically very structured, and
reprojectingthepixelsintonovelcamerapo-
theyarealsooneofthemostcommontypesofenvironment
sitions. On the left is the result of virtually
to model. The goal of the research was to model architec-
moving one meter forward, on the right is
tureinawaythatisconvenient,requiresrelativelyfewpho-
theresultofvirtuallymovingonemeterback-
tographs, and produces freely navigable and photorealistic
ward. Note the dark disoccluded areas pro-
models.
duced by these virtual camera moves; these
TheproductofthisresearchwasFac¸ade[8],aninterac-
areaswerenotseenintheoriginalstereopair.
tivecomputerprogramthatenablesausertobuildphotore-
In the Immersion ’94animations (available at
alisticarchitecturalmodelsfromasmallsetofphotographs.
http://www.debevec.org/Immersion,thesere-
InFac¸ade,theuserbuildsa3Dmodelofthescenebyspec-
gionswereautomaticallyfilledinfromneigh-
ifying a collection of geometric primitives such as boxes,
boringstereopairs.
arches,andsurfacesofrevolution. However,unlikeinatra-
ditionalmodelingprogram,theuserdoesnotneedtospec-
ifythedimensionsorthelocationsofthesepieces. Instead,
the user corresponds edges in the model to edges marked
inthephotographs, andthecomputerusesanoptimization Fac¸ade to generate photorealistic views of the scene by
processtodeterminetheshapesandpositionsoftheprimi- usingalloftheavailablephotographs.
tivesthatmakethemodelagreewiththephotographedge- Some additional research done in the context of the
ometry(Fig. 2). Fac¸adesystemenablesthecomputertoautomaticallyrefine
a basic recovered model to conform to more complicated
architecturalgeometry. Thetechnique,calledmodel-based
stereo, displaces the surfaces of the model to make them
maximally consistent with their appearance across multi-
ple photographs. Thus, a user can model a bumpy wall
as a flat surface, and the computer will compute the re-
lief. This technique was employed in modeling the West
fac¸adeofthegothicRouencathedralfortheinteractiveart
installation Rouen Revisited shown at the SIGGRAPH 96
art show. Most of the area between the two main towers
seen in Fig. 3 was originally modeled as a single poly-
gon. TheRouenprojectalsomotivatedtheadditionofnew
features to Fac¸ade to solve for unknown focal lengths and
centers of projection in order to make use of historic pho-
tographsofthecathedral.
Figure 2. A screen snapshot from Fac¸ade. The
windows include the image viewers at the
left, where the user marks architectural
edge features, and model viewers, where
the user instantiates geometric primitives
(blocks) and corresponds model edges to
image features. Fac¸ade’s reconstruction
feature then determines the camera param-
eters and position and dimensions of all Rendering: 1996 Rendering: 1896 Rendering: painting
the blocks that make the model conform to
the photographs. The other windows in-
Figure 3. Rouen Revisited. Synthetic views
clude the toolbar, the camera parameter di-
of the Rouen cathedral from the Rouen Revis-
alog, the block parameter/constraint dialog,
ited art installation. Left: a synthetic view
and the main image list window. See also
created from photographs taken in January,
http://www.debevec.org/Thesis/.
1996. Middle: a synthetic view created from
historic postcards showing the cathedral at
the time Monet executed his series of paint-
Fac¸ade solves directly for the architectural dimensions
ings (1892-1894). Right: a synthetic view of
ofthescene: thelengthsofwalls,thewidthsofdoors,and
one of Monet’s twenty-eight paintings of the
theheightsofroofs,ratherthanthemultitudeofvertexco-
cathedral projected onto its historic geome-
ordinatesthatastandardphotogrammetricapproachwould
try,renderingitfromanovelviewpoint.
try to recover. As a result, the reconstruction problem be-
comes simpler by orders of magnitude, both in computa-
tionalcomplexityandinthenumberofimagefeaturesthatit
isnecessaryfortheusertomark. Thetechniquealsoallows
4 The Campanile Movie: Rendering in Real
theusertoexploitarchitecturalsymmetries–modelingre-
Time
peatedstructuresandcomputingredundantdimensionsonly
once.
Like any structure-from-multiple-views algorithm, The Campanile Movie project aimed to use Fac¸ade to
Fac¸ade solves for where the original cameras were in the reconstruct a model of the UC Berkeley bell tower and its
scene. With the camera positions known, any one of the entire surrounding campus - to increase the photorealism
photographs can be projected back onto the reconstructed of the renderings by showing both a building and its con-
geometry using projective texture mapping, which allows text. Fordatacapture,Itookphotographsfromtheground,
fromthetower,and(thankstoBerkeleyprofessorofarchi-
tectureCrisBenton)fromabovethetowerusingakite. The
finalmodelwebuiltinFac¸adecontainedfortyofthecam-
pusbuildings; thebuildingsfurtherawayappearedonlyas
texturesprojectedontotheground. Therewereafewthou-
sand polygons in the model, and the sixteen images (Fig.
4) used in rendering the scene fit precisely into the avail-
abletexturememoryoftheSiliconGraphicsRealityEngine
we used for rendering. Using OpenGL and a hardware-
accelerated view-dependent texture-mapping technique –
selectively blending between the original photographs de-
pending on the user’s viewpoint [9] – made it possible to
renderthesceneconvincinglyinrealtime.
(a)
5 Fiat Lux: Capturing Lighting and Adding
Objects
We most recently used Fac¸ade to model and render the
interior of St. Peter’s Basilica for the animation Fiat Lux
(Fig.5),whichwasshownattheSIGGRAPH99Electronic
Theater. InFiatLux,ourgoalwastonotonlycreatevirtual
renderings of moving through St. Peter’s, but to augment
thespacewithanimatedcomputer-generatedobjectsinthe
serviceofanabstractinterpretationoftheconflictbetween (b)
Galileoandthechurch.
The key to making the computer-generated objects ap-
pear to be truly present in the scene was to illuminate the
CG objects with the actual illumination from the Basilica.
Torecordtheilluminationweusedahighdynamicphotog-
raphymethod[7]wehaddevelopedinwhichaseriesofpic-
turestakenwithdifferingexposuresarecombinedintoara-
dianceimage–withoutthetechnique,camerasdonothave (c) (d)
nearly the range of brightness values to accurately record
the full range of illumination in the real world. We then Figure 4. TheCampanileMovie. At top are the
usedanimage-basedlighting[4]techniquetoilluminatethe original sixteen photographs used for ren-
CG objects with the images of real light using a global il- dering; four additional aerial photographs
lumination rendering system. In addition, we used an in- were used in modeling the campus geome-
verseglobalillumination[17]techniquetoderivelighting- try. In the middle is a rendering of the cam-
independentreflectancepropertiesofthefloorofSt.Peter’s, pus buildings reconstructed from the pho-
allowingtheobjectstocastshadowsonandappearinreflec- tographs using Fac¸ade; the final model also
tionsinthefloor. Havingthefullrangeofilluminationwas included photogrammetrically recovered ter-
additionallyusefulinproducingavarietyofrealisticeffects rain extending out to the horizon. At bottom
ofcinematography,suchassoftfocus,glare,vignetting,and are two computer renderings of the Berke-
lensflare. ley campus model obtained through view-
dependent texture mapping from the SIG-
6 The Light Stage: A Photometric Approach GRAPH 97 animation. The film can be seen
athttp://www.debevec.org/Campanile/.
toDigitizingCulturalArtifacts
Thebesttechniquespresentedtodateformodelingreal-
world artifacts involve two stages. First, the geometry of
the artifact is measured using a range scanning device and
somesortofmeshmergingalgorithm[13,16,12]. Second,
photographstakenundercontrolledlightingconditionsare
analyzedtodeterminethecolorofeachpointontheobject’s
surface to produce texture maps for the object. A notable
exception to this process is the trichromatic laser scanner
device presented in [3, 2] which is able to measure object
surfacecolorinthesameprocessasobtainingthegeomet-
ric laser scan. All of these techniques have produced 3D
models with accurate models of surface geometry as well
aslighting-independentdiffusetexturemaps.
These techniques represent significant advances in the
(a)
fieldofartifactdigitization, andhaveshownthemselvesto
work particularly well on objects with close to Lamber-
tian reflectance properties such as aged marble sculptures.
However, these existing techniques for model acquisition
are difficult to apply to a large class of artifacts exhibit-
ing spatially varying specular reflection properties, com-
plex BRDFs, translucency, and subsurface scattering. As
a result, artifacts featuring polished silver, gold, glass, fur,
cloth,jewels,jade,leaves,orfeathersremainverychalleng-
ingtoaccuratelydigitizeandtoconvincinglyrender. Since
(b) a large class of cultural artifacts are specifically designed
tohaveintricategeometryandtoreflectlightininteresting
ways,developingnewtechniquesforsuchobjectsisanim-
portantavenueforresearch.
OurrecentworkinvolvingtheLightStage[5,11](Figure
6) has explored a technique for creating re-lightable com-
putermodelsofartifactsandpeoplewithoutexplicitlymod-
elingtheirgeometricorreflectanceproperties. Instead,the
artifact is illuminated from a dense array of incident illu-
minationdirectionsandasetofdigitalimagesaretakento
recordtheartifact’sreflectancefield,afunctionthatrecords
how the object would appear seen from any angle as illu-
minated by any lighting environment. The data within a
reflectancefieldcanbequicklycombinedtogetherinorder
to produce images of an artifact under any form of illumi-
(c) nation, including lighting environments captured from the
realworld.
Figure 5. Fiat Lux. The animation Fiat Lux In[11],ourgroupusedanimprovedlightstagedeviceto
shown at the SIGGRAPH 99 Electronic The- digitize several Native American cultural artifacts exhibit-
ater used Fac¸ade [8] to model and render ingmanyofthegeometricandreflectancepropertieswhich
the interior of St. Peter’s Basilica from sin- have traditionally been difficult to model and render. We
gle panorama assembled from a set of ten furthermore described an interactive lighting tool that al-
images. Each image was acquired using lowsartifactstobere-illuminatedbyauserinrealtime,and
highdynamicrangephotography[7],inwhich proposed image-based rendering techniques that allow an
each image is taken with a range of differ- artifacttobemanipulatedin3Daswellasbeingarbitrarily
ent exposure settings and then assembled illuminated. An advantage of this technique for capturing
into a single image that represents the full andrenderingobjectsisthattheobjectneednothavewell-
range of illumination in the scene. The re- definedsurfacesoreasytomodelreflectanceproperties;the
covered lighting was used to illuminate the object can have arbitrary translucency, self-shadowing, in-
synthetic objects added to the scene, giving terreflection, subsurface scattering, and fine surface detail.
them the correct shading, shadows, reflec- This is helpful for modeling and rendering subjects such
tions,andhighlights. Thefilmiscanbeseen as human faces, jewelry, and clothing which exhibit all of
athttp://www.debevec.org/FiatLux/. theseproperties.
The significant drawback to this approach is the large
amountofdatathatneedstobecapturedandstoredforthe
object: on the order of at least a thousand lighting direc-
tionsforeachobjectviewingdirection.Asaresult,wehave
yet to capture a model with both a dense array of viewing
andlightingdirections,sothechangesinviewpointwehave
beenabletoobtainthusfarhavebeenlimited.
A recent project at MIT and MERL [14] used a device
similar to our light stage to capture several objects from
moderatelydensearraysofviewingdirectionsandapprox-
imately 64 lighting directions, and used a silhouette-based
volumetricintersectiontechniquetoobtainroughgeometry
forthedigitizedobjects. AswesawearlierwithourFac¸ade
work,usingbasicscenegeometryasastructureforproject-
ingimagesallowssparselysampledviewingdirectionstobe
effectively extrapolated to new viewing directions, at least
(a)
forrelativelydiffuseobjectswherethelightreflectingfrom
a surface remains nearly constant under changing viewing
directions. Something that still has not been leveraged in
the digitizing process is that the way a particular point on
anobjectsurfacereflectslight(thepoint’sfour-dimensional
BRDF)canoftenbewell-approximatedbyalow-parameter
reflectance model encompassing, for example, diffuse and
specular parameters. The next project I will describe ex-
ploitsthischaracteristic.
(b)
7.LinearLightSourceReflectometry
Animportantaspectofmanyculturalartifactsisthatthey
havespatially-varyingreflectanceproperties:theyaremade
ofdifferentmaterialseachofwhichreflectlightinadiffer-
entway.Anexampleisanilluminatedmanuscript,inwhich
differentcolorsofinksanddifferenttypesofpreciousmet-
(c) (d) alsareappliedtothesurfaceofparchmenttoproducetex-
tual designs and illustrations. We have found the case of
Figure 6. Digitizing objects in the Light Stage. digitizinganilluminatedmanuscripttobeparticularlygood
The Light Stage [5, 11] is designed to illumi- for developing algorithms to digitize spatially-varying re-
nateanartifactfromeverydirectionlightcan flectance properties since the geometry of a manuscript is
come from in a short period of time. In (a), a relativelyflat,makingthegeometryrecoveryandreflectom-
NativeAmericanheaddressisbeingdigitized. etryprocessessimpler.
This allows a digital video camera to directly Digitizing the manuscript within the light stage would
capture the subject’s reflectance field – how it bepossible, butsincethemanuscriptcontainsbothdiffuse
transforms incident illumination into radiant and sharply specular regions it would need to be illumi-
illumination – as an array of images (b). As natedbyalargenumberoflightingandviewingdirections
a result, we can then synthetically illuminate to be faithfully captured. To avoid this problem, we de-
thesubjectunderanyformofcomplexillumi- velopedanewdigitizingtechniquedescribedin[10]based
nation directly from this captured data. The on using a translating linear light source as the illuminant
Native American headdress is shown as if il- (see Fig. 7). We found that in passing the light source
luminated by light from a cathedral (c) and a over the artifact and viewing it from an oblique angle, we
forest(d). couldindependentlyobservethediffuseandspecularreflec-
tion of the light at each surface point. To turn these ob-
servations into digitized surface reflectance properties, we
fit the observed diffuse and specular reflections to a stan-
dard physically-motivated reflectance model with diffuse, artifact: a dataset of how a environment or artifact would
specular,andspecularroughnesscoefficients. Furthermore, look from any angle under any form of incident illumina-
we found that by passing the light source across the ob- tion. The problems of digitizing environments and objects
ject twice at different angles, we can estimate surface nor- are related but have differences. Environments are larger,
malsateverypointontheobjectaswell. Simpleextensions andoftenmoregeometricallycomplex,makingthemharder
to the technique – adding a laser stripe and a back light – to photograph from a dense sampling of all possible an-
makeitpossibletocaptureper-pixelsurfaceheightaswell gles. Environmentsarealsohardertoilluminatewithcon-
asper-pixelsurfacetranslucencyinadditiontotheotherre- trolledlighting-outdoorscenesaremosteasilyilluminated
flectanceparameters. by natural lighting rather than controlled sources, and in-
door scenes tend to exhibit a great deal of mutual illumi-
nationbetweensurfaces. Asaresult,manycomputermod-
els which have been made of real-world environments are
lighting-dependent,showingthesceneunderjustoneform
ofillumination, suchasintheBanffforest, Berkeleycam-
pus, and St. Peter’s Basilica models. Nonetheless, these
modelscanbeveryeffectivedigitizations,perhapsbecause
for environments we are able to consider the illumination
conditionstobepartoftheenvironment.
For digitizing artifacts, being able to capture their
lighting-independent reflectance properties is much more
important. Partofthereasonisthatwhenweviewavirtual
artifactfromdifferentangles,wemostnaturallyassumethat
thelightingstaysinourownframeofreferenceratherthan
rotating with the artifact. Also, in order to place a virtual
(a)
artifact into a virtual museum environment, it is necessary
to illuminate the artifact with the light present in the ap-
propriate part of the museum. Fortunately, artifacts are in
many respects easier to digitize than environments. They
are smaller, and thus can be brought into object scanning
facilitiestobephotographedfrommanyanglesandillumi-
nated under controlled forms of illumination. As a result,
most artifact modeling projects, such as the Native Amer-
ican Artifacts and the illuminated manuscript, have been
successfulatrecoveringlighting-independentmodelsofthe
(b) (c)
artifacts.However,thenativeartifactscouldonlybeviewed
from a limited set of viewing angles, and the manuscript,
Figure 7. Linear Light Source Reflectometry.
being a flat object, could be digitized without concern for
The spatially-varying surface reflectance pa-
self-occlusionandmutualilluminationeffects.
rameters of an illuminated manuscript are
recordedusingasimpleapparatusbasedon
9.Conclusion
atranslatinglinearlightsource. Theper-pixel
estimatesofdiffuseandspecularparameters,
as well as surface normals, can be used to Theprojectssurveyedinthispapereachpresentadiffer-
createphotorealreal-timerenderings(below) entapproachtodigitizingartifactsandenvironments.While
of the artifact from any viewing angle and in each aims to produce photoreal results, the methods sig-
anylightingenvironment. nificantly differ in the nature of the measurements taken,
the amount of user input required, the amount of geomet-
ric detail recovered, and the degree to which the models
canbeseenundernovellightingandfromnewviewpoints.
8.Discussion Methods for environment and object digitizing in the fu-
ture will likely draw upon several of these techniques to
Lookingattheseprojectstogether,wecanseethateach improve their results, though techniques for digitizing ob-
one of them can be characterized as acquiring or estimat- jects as compared to environments will likely continue to
ing a portion of the reflectance field of an environment or besignificantlydifferent.
Acknowledgements [10] A.Gardner,C.Tchou,T.Hawkins,andP.Debevec. Linear
light source reflectometry. In Proceedings of SIGGRAPH
2003,ComputerGraphicsProceedings,AnnualConference
Theprojectsdescribedinthispaperarejointworkwith
Series,July2003.
Michael Naimark, John Woodfill, Leo Villareal, Golan
[11] T.Hawkins, J.Cohen, andP.Debevec. Aphotometricap-
Levin, C.J. Taylor, Jitendra Malik, George Borshukov,
proach to digitizing cultural artifacts. In Proc. 2nd Inter-
YizhouYu, TimHawkins, JonathanCohen, AndrewGard-
national Symposium on Virtual Reality, Archaeology, and
ner, and Chris Tchou. Abjiheet Ghosh and Chris Tchou CulturalHeritage(VAST2001),pages333–342,December
prepared the illuminated manuscript renderings in Fig. 7, 2001.
andMayaMartinezarrangedfortheuseoftheculturalar- [12] M.Levoy,K.Pulli,B.Curless,S.Rusinkiewicz,D.Koller,
tifacts used in this work. This work has been funded by L.Pereira,M.Ginzton,S.Anderson,J.Davis,J.Ginsberg,
IntervalResearchCorporation,theNationalScienceFoun- J.Shade,andD.Fulk. Thedigitalmichelangeloproject:3d
scanningoflargestatues. ProceedingsofSIGGRAPH2000,
dation,theCaliforniaMICROprogram,InteractivePictures
pages131–144,July2000.
Corporation,JSEPcontractF49620-93-C-0014,U.S.Army
[13] S. Marschner. Inverse Rendering for Computer Graphics.
contractDAAD19-99-D-0046,andtheUniversityofSouth-
PhDthesis,CornellUniversity,August1998.
ern California. The content of this information does not
[14] W. Matusik, H. Pfister, A. Ngan, P. Beardsley, R. Ziegler,
necessarilyreflectthepositionorpolicyofthesponsorsand andL.McMillan. Image-based3dphotographyusingopac-
noofficialendorsementshouldbeinferred. ityhulls. ACMTransactionsonGraphics, 21(3):427–437,
July2002.
[15] M.Naimark,J.Woodfill,P.Debevec,andL.Villareal. Im-
References
mersion’94. http://www.debevec.org/Immersion/,1994.
[16] H.Rushmeier, F.Bernardini, J.Mittleman, andG.Taubin.
[1] A.C.Addison,D.MacLeod,G.Margolis,M.Naimark,and Acquiringinputforrenderingatappropriatelevelsofdetail:
H.-P. Schwartz. Museums without walls: New media for Digitizingapieta`.EurographicsRenderingWorkshop1998,
newmuseums. InR.Cook,editor,ComputerGraphicsan- pages81–92,June1998.
nual Conference Series (SIGGRAPH 95), pages 480–481, [17] Y.Yu,P.Debevec,J.Malik,andT.Hawkins. Inverseglobal
August1995. illumination: Recoveringreflectancemodelsofrealscenes
[2] R.Baribeau,L.Cournoyer,G.Godin,andM.Rioux.Colour fromphotographs. InSIGGRAPH99,August1999.
[18] R.ZabihandJ.Woodfill. Non-parametriclocaltransforms
three-dimensional modelling of museum objects. Imaging
forcomputingvisualcorrespondence. InEuropeanConfer-
thePast,ElectronicImagingandComputerGraphicsinMu-
enceonComputerVision,pages151–158,May1994.
seumandArchaeology,pages199–209,1996.
[3] R. Baribeau, M. Rioux, and G. Godin. Color reflectance
modelingusingapolychromaticlasersensor. IEEETrans.
PatternAnal.MachineIntell.,14(2):263–269,1992.
[4] P. Debevec. Rendering synthetic objects into real scenes:
Bridging traditional and image-based graphics with global
illuminationandhighdynamicrangephotography. InSIG-
GRAPH98,July1998.
[5] P.Debevec,T.Hawkins,C.Tchou,H.-P.Duiker,W.Sarokin,
and M. Sagar. Acquiring the reflectance field of a human
face.ProceedingsofSIGGRAPH2000,pages145–156,July
2000.
[6] P. E. Debevec. Modeling and Rendering Architecture
fromPhotographs. PhDthesis, UniversityofCaliforniaat
Berkeley,ComputerScienceDivision,BerkeleyCA,1996.
http://www.debevec.org/Thesis.
[7] P.E.DebevecandJ.Malik.Recoveringhighdynamicrange
radiancemapsfromphotographs. InSIGGRAPH97,pages
369–378,August1997.
[8] P.E.Debevec,C.J.Taylor,andJ.Malik.Modelingandren-
deringarchitecturefromphotographs: Ahybridgeometry-
andimage-basedapproach.InSIGGRAPH96,pages11–20,
August1996.
[9] P.E.Debevec,Y.Yu,andG.D.Borshukov. Efficientview-
dependent image-based rendering with projective texture-
mapping. In 9th Eurographics workshop on Rendering,
pages105–116,June1998.